Non-volatile memories provide long-term storage of data. More particularly, non-volatile memories can retain the stored data even when not powered. Magnetic (rotating) hard disk drives (HDD) dominate this storage medium due to lower cost compared to solid state disks (SSD). Optical (rotating) disks, tape drives and others have a smaller role in long-term storage systems. SSDs are preferred for their superior performance (fast access time), mechanical reliability and ruggedness, and portability. Flash memory, more specifically NAND flash, is the dominant SSD medium today.
RRAM, PCM, MAGRAM and others, will likely play a larger role in the future, each of them having their own advantages and disadvantages. They may ultimately replace flash memories, initially for use as a “write buffer” and later to replace “SLC flash” and “MLC flash.” MLC NAND flash is a flash memory technology using multiple levels per cell to allow more bits to be stored using the same number of transistors. In SLC NAND flash technology, each cell can exist in one of two states, storing one bit of information per cell. Most MLC NAND flash memory has four possible states per cell, so it can store two bits of information per cell.
These semiconductor technology driven “flash alternatives,” i.e., RRAM, PCM, MAGRAM and others, have several advantages over any (SLC or MLC) flash because they: 1) allow data to be written over existing data (without prior erase of existing data), 2) allow for an erase of individual bytes or pages (instead of having to erase an entire block), and 3) possess superior endurance (1,000,000 write-erase cycles compared to typical 100,000 cycles for SLC flash and less than 10,000 cycles for MLC flash).
HDDs have several platters. Each platter contains 250-5,000 tracks (concentric circles). Each track contains 64 to 256 sectors. Each sector contains 512 bytes of data and has a unique “physical (memory) address.” A plurality of sectors is typically combined to form a “logical block” having a unique “logical address.” This logical address is the address at which the logical block of physical sectors appears to reside from the perspective of an executing application program. The size of each logical block and its logical address (and/or address ranges/boundaries) is optimized for the particular operating system (OS) and software applications executed by the host processor. A computer OS organizes data as “files.” Each file may be located (stored) in either a single logical block or a plurality of logical blocks, and therefore, the location of files typically traverses the boundaries of individual (physical) sectors. Sometimes, a plurality of files has to be combined and/or modified, which poses an enormous challenge for the memory controller device of a non-volatile memory system.
SSDs are slowly encroaching on the HDD space and the vast majority of NAND flash in enterprise servers utilizes a SLC architecture, which further comprises a NAND flash controller and a flash translation layer (FTL). NAND flash devices are generally fragmented into a number of identically sized blocks, each of which is further segmented into some number of pages. It should be noted that asymmetrical block sizes, as well as page sizes, are also acceptable within a device or a module containing devices. For example, a block may comprise 32 to 64 pages, each of which incorporates 2-4 Kbit of memory. In addition, the process of writing data to a NAND flash memory device is complicated by the fact that, during normal operation of, for example, single-level storage (SLC), erased bits (usually all bits in a block with the value of ‘1’) can only be changed to the opposite state (usually ‘0’) once before the entire block must be erased. Blocks can only be erased in their entirety, and, when erased, are usually written to ‘1’ bits. However, if an erased block is already there, and if the addresses (block, page, etc.) are allowed, data can be written immediately; if not, a block has to be erased before it can be written to.
FTL is the driver that works in conjunction with an existing operating system (or, in some embedded applications, as the operating system) to make linear flash memory appear to the system like a disk drive, i.e., it emulates a HDD. This is achieved by creating “virtual” small blocks of data, or sectors, out of flash's large erase blocks and managing data on the flash so that it appears to be “write in place” when in fact it is being stored in different locations in the flash. FTL further manages the flash so that there are clean/erased places to store data.
Given the limited number of writes that individual blocks within flash devices can tolerate, wear leveling algorithms are used within the flash devices (as firmware commonly known as FTL or managed by a controller) to attempt to ensure that “hot” blocks, i.e., blocks that are frequently written, are not rendered unusable much faster than other blocks. This task is usually performed within a flash translation layer. In most cases, the controller maintains a lookup table to translate the memory array physical block address (PBA) to the logical block address (LBA) used by the host system. The controller's wear-leveling algorithm determines which physical block to use each time data is programmed, eliminating the relevance of the physical location of data and enabling data to be stored anywhere within the memory array and thus prolonging the service life of the flash memory. Depending on the wear-leveling method used, the controller typically either writes to the available erased block with the lowest erase count (dynamic wear leveling); or it selects an available target block with the lowest overall erase count, erases the block if necessary, writes new data to the block, and ensures that blocks of static data are moved when their block erase count is below a certain threshold (static wear leveling).
MLC NAND flash SSDs are slowly replacing and/or coexisting with SLC NAND flash in newer SSD systems. MLC allows a single cell to store multiple bits, and accordingly, to assume more than two values; i.e., ‘0’ or ‘1’. Most MLC NAND flash architectures allow up to four (4) values per cell; i.e., ‘00’, ‘01’, ‘10’, or ‘11’. Generally, MLC NAND flash enjoys greater density than SLC NAND flash, at the cost of a decrease in access speed and lifetime (endurance). It should be noted, however, that even SLC NAND flash has a considerably lower lifetime (endurance) than rotating magnetic media (e.g., HDDs), being able to withstand only between 50,000 and 100,000 writes, and MLC NAND flash has a much lower lifetime (endurance) than SLC NAND flash, being able to withstand only between 3,000 and 10,000 writes. As is well known in the art, any “write” or “program” to a block in NAND flash (floating gate) requires an “erase” (of a block) before “write.”
Despite its limitations, there are a number of applications that lend themselves to the use of MLC flash. Generally, MLC flash is used in applications where data is read many times (but written few times) and physical size is an issue. For example, flash memory cards for use in digital cameras would be a good application of MLC flash, as MLC can provide higher density memory at lower cost than SLC memory.
When a non-volatile storage system combines HDD, SLC and MLC (setting aside volatile memory for buffering, caching etc) in a single (hybrid) system, new improvements and solutions are required to manage the methods of writing data optimally for improved life time (endurance) of flash memory. Accordingly, various embodiments of a NAND flash storage system that provides long lifetime (endurance) storage at low cost are described herein.
The following description is presented to enable one of ordinary skill in the art to make and use the disclosure and is provided in the context of a patent application and its requirements. Various modifications to the preferred embodiment and the generic principles and features described herein will be readily apparent to those skilled in the art. Thus, the present disclosure is not intended to be limited to the embodiments shown, but is to be accorded the widest scope consistent with the principles and features described herein.